AI Text Detection Fails to Settle Goncourt Novel's Fate After €300K Deal

For AI and machine learning professionals, this controversy is less about literary merit than about the measurable limits of current AI text detection. The Académie Goncourt acknowledged that detection tools disagree, leaving only a hedged conclusion that the novel was 'very largely' AI-produced. The case is a live test of AI provenance, attribution uncertainty, and bias in automated content policing.

The Prix Goncourt, France's most prestigious literary prize, jolted the publishing world on Sept. 25 when its academy removed Thelyson Orelien's debut novel 'It Was Either That or Die' from the 2026 longlist after online accusations that the book was written with artificial intelligence. The decision came after a little-known X account, 'Balance ton Claude,' posted the AI allegations; essayist Samuel Fitoussi later claimed ownership of the account, saying he operated it with other unidentified individuals. The Académie Goncourt said the removal was necessary 'to preserve the integrity of the Prize and that of the Academie, whose central role is to promote and honour literature written by women and men.' Orelien, a 38-year-old Haitian-Canadian writer, denied using AI and described the campaign against him as racially motivated and an attempt to silence his voice. He promised to produce a file proving his authorship. The stakes are substantial: the novel had been sold in more than 20 countries, including the United States, and Paris publisher Grasset reportedly paid 300,000 euros for French rights.

“

Radio-Canada separately found that a novella by Orelien that won an audience prize in 2010 contained phrases taken from 'Le matin des magiciens,' a book by Louis Pauwels and Jacques Bergier published in 1960.

The case exposes a brittle technical reality: AI text detection is not settled science. The academy itself said 'several credible researchers and journalists' had concluded the novel was 'in all likelihood very largely' produced by AI, yet it also acknowledged that different detection tools produced different results. That hedging is important. Natural language generation has advanced to the point where stylistic analysis, perplexity scoring, and detector classifiers frequently disagree, especially for literary prose that may be heavily edited or translated. For a prize that claims to honor human creation, the absence of a reliable, transparent method to verify authorship raises uncomfortable questions about the burden of proof, the risk of false accusations, and the potential for reputational damage based on unverifiable machine outputs.

The AI allegations did not stand alone. Within days, Canadian media added plagiarism claims from Orelien's earlier work. Montreal's La Presse reported that a newspaper column he wrote in 2013 was almost entirely a lightly edited translation of an article published several days earlier by The Philippine Star. Radio-Canada separately found that a novella by Orelien that won an audience prize in 2010 contained phrases taken from 'Le matin des magiciens,' a book by Louis Pauwels and Jacques Bergier published in 1960. Source reporting differs slightly on whether the short story was from 2012 or 2010, but the pattern is the same: a history of alleged uncredited borrowing. The plagiarism reports complicate the AI narrative. If the author has reused others' words for years, the AI detection question may become one part of a larger integrity problem. Yet Orelien had not responded to the plagiarism accusations at the time of reporting, while his French and Canadian publishers initially offered full support over the AI claims but did not respond to later requests for comment.